THE BOOK--Playing The Percentages In Baseball

pos1 pos2 r1 r2 diff G
4 5 -4.7 -1.3 3.4 4860
4 6 0.5 -4.1 -4.6 5393
5 6 0.4 -4 -4.4 3620

Not as many games, and the skillsets don’t necessarily translate the process isn’t as strong here. From the perspective of the SS, guys who move there from 2B or 3B do equally poorly (by 4.5 runs).  However, guys who move between 2B and 3B do better at 3B, meaning that 2B as a group are better.  Splitting the difference, we best-fit as:

4.3 SS
1.7 2B
0.0 3B

***

When we look at 1B with the other 3 IF positions, we get:
pos1 pos2 r1 r2 diff G
3 4 -2 -4.2 -2.2 1191
3 5 -1.3 -7.4 -6.1 3375
3 6 0.6 -9.6 -10.2 963

We see that they do better at 2B than 3B, which gives us reason to again put the 2B and 3B at the same level.  Anyway, compared to the 3 IF positions as a group, 1B is around 6 runs worse.

Clearly only good fielders at 1B will play at other positions, so we can see how if you looked at all 1B that they’d be more exposed in the IF.  How much more is a tough call.

***

Similarly, here are the 3 OF positions compared to 1B:
pos1 pos2 r1 r2 diff G
3 7 -1 -5 -4 2879
3 8 1.5 -4 -5.5 656
3 9 -2.4 -10.5 -8.1 2338

Once again, around a 6 run difference, with the same skepticism as above.

***

We see from the standpoint of the 1B that they are as far away from the 3 IF positions as the 3 OF positions.  Of course, we also have the handedness issue.

Putting it all together, we get something like this:
3 SS
-1 2B
-2 3B

8 CF
-4 LF
-4 RF

-6 1B

The relationship among the 3 IF hold, and among the 3 OF hold.  The average for the 3 IF and the 3 IF is 0, which are both 6 runs from the 1B.  If I throw a blanket +6 for the C, we get:
8 CF
6 C
3 SS
-1 2B
-2 3B
-4 LF
-4 RF
-6 1B

The handedness thing is an issue for sure.  We figured in another thread, that the bonus would be +3 for the IF positions, and -3 for the OF positions.  So that gives us:

6 C
6 SS
5 CF
2 2B
1 3B
-6 1B
-7 LF
-7 RF

Obviously, 1B is now out-of-whack here.  If we move things up and down a little, we end up with:

12.5 C
7.5 SS
2.5 CF
2.5 2B
2.5 3B
-7.5 LF
-7.5 RF
-12.5 1B

Obviously, I took a few liberties, especially with catcher and firstbase.  The rest hold pretty well.  The relationship between SS and 2B/3B hold as original.  The same applies for CF and LF/RF.  Perhaps 1B should really be -10 and C should be +10.

Anyway, that’s pretty much it.

***

Oh, let me look at the actual IF/OF movement.  SS playing in the OF are +8.5 runs ahead of the three OF.  That’s a match to the third-to-last chart (the one before the handedness adjustment).

2B who played in the three OF positions are around 1 run behind the OF.  This is also a match to the third-to-last chart.  3B are 2 runs behind the 3 OF positions, and again, another match.

So, we’ve got some pretty strong agreement here.

I took the 30 players with the most games at each position, each year.  That gives me 180 seasons for each position.  Here is how the regulars did at each position, per 150 games:

pos Rper150_weighted class
3 1.1 1_Regulars
4 1.4 1_Regulars
5 1.6 1_Regulars
6 1.4 1_Regulars
7 -1.1 1_Regulars
8 2.1 1_Regulars
9 0.7 1_Regulars

As you can see, regulars are a bit above average fielders.  Except in LF.

I then took the next 30 players who played the most games (players 31 through 60) at each position, each year.  Here’s how the backups did:

pos Rper150_weighted class
3 -2.8 2_Backups
4 -0.7 2_Backups
5 -3.0 2_Backups
6 -3.2 2_Backups
7 +4.4 2_Backups
8 -5.4 2_Backups
9 -1.4 2_Backups

As you can see, backup LF are very good fielders.  Otherwise, backup fielders are below average.

Finally, the rest of the players (players 61st, onwards in terms of games played per position):

pos Rper150_weighted class
3 -5.4 3_Rest
4 -9.9 3_Rest
5 -8.7 3_Rest
6 -14.8 3_Rest
7 -2.9 3_Rest
8 -5.3 3_Rest
9 -2.8 3_Rest

The infielders really suck, especially SS.  The 1B/OF are bad too.

Great stuff!  Lots to digest, but excellent stuff.  Definitely an area that needs more looking into, as we can infer from the fact that it takes a lot of tweaking to get everything to line up.

Any speculation as to why in LF, the backups are substantially better than the regulars in fielding?  We can say that it is because lots of the backups are defensive replacements, but that applies to some extent at other positions as well.  The only thing I can think of is that all other positions pretty much have good defenders as the regulars.  LF is a place where you stick a lousy fielder that for whatever reason does not play first base.  You can say the same about RF, but I think RF’ers have to have good arms, so you don’t just stick anyone with a good bat in RF.  Plus even if you want to replace your RF late in a game that you are winning, you can’t put a slick fielder there if he does not have a good arm.  In LF, you can pretty much put any slick utility fielder (like Bruntlett who comes in for Burrell) out there and you don’t need a good OF arm.

Someone suggested that LF is the DH for the NL.

Also, with the number of opposite field hits these past few years, better fielders are at RF than LF (and the arm of course).  So, there’s this shift.  And, if LF is a dumping ground for good hit, no field for the regulars, you have no choice but to think his backup will be the opposite.

I think that there is no such thing as a backup left fielder, just a backup outfielder that plays all three spots. (Okay, so I’m oversimplifying here.) If left field is easier than right field and right field is easier than center, your typical fourth outfielder should be better in left than he is in right, and better in right than he is in center.

So I think that for the backups table, you’re looking at essentially the same player pool at all three outfield positions.

That may also explain why the backup second baseman is the closest to being an average defender at his position - because oftentimes he’s the same guy as the backup shortstop.

I have no doubt that you are correct, as I espouse the same thing.

Players are put in roles, not positions.  You are either a C, IF, OF, or everything else.  That’s the fact of it.  If you can’t cut it in one, you move on to the next one.

Tom,

Using matched pairs, I am looking at comparing the inputs to UZR for players that play at various positions.  I was started by comparing UZR to help find a procedure/path for comparing individual factors as a whole.

Our overall numbers match up pretty good.  I have RF fielding a little better than LF, but almost everything matches, except for 1B/3B.  With your model you have the difference being a difference of 15, I was only able to get a difference of 3.7.  A difference of 11.3 is just too much. 

The individual factors are
3B to 1B increase
DP = .6, Range = .6 and error = 2.5.

Thanks for your time.

If anyone, including Tango, is doing a follow-up, the aging thing is important, especially for players that jump a few positions in the spectrum. 

For example, going from 3B to 1B, it is likely that you were much younger when playing 3B, which is going to give the illusion of a smaller gap without controlling for age. Then again, there is some evidence that you need experience at 1B, so maybe that is a bad example.  Not to mention the fact that players who move left tend to be those who don’t age very well, defensively.

There are also selective sampling issues which would make the gap look larger.  If you are bad at SS, which includes a little bad luck, you are more likely to get moved to 2B or 3B, making the gap look larger than it really is (a true 0 at SS plays as a -2 because he got a little unlucky and is then a +4 at 2B - his true gap is 4 runs, but it looks like his gap is 6 runs based on actual data).

I would do a follow-up on this if I could get help on the age adjustments MGL discusses. I’d use the UZR from 2002-2008. Also, I am interested in using this method historically. Actually that is the main reason why I would do it to begin with. It wouldn’t be too difficult to devise a decent and simple non play-by-play fielding metric. For infielders, calculate expected assists and expected DP’s, and adjust for LHP/RHP pitching and DP opportunities. For outfielders, calculate expected putouts while adjusting for LHP/RHP pitching, and give them credit for their assists. I would assign run values based on PBP data from retrosheet, and then adjust the run-values according to the run-environment. For instance, an IF assist would be worth more in 1930 than in 1908. And I already have defensive innings estimates for all players in the BDB database.

Tango already compared different positions using 2002-2008 UZR:

http://www.insidethebook.com/ee/index.php/site/comments/dual_positions_using_buzr/#comments

Nonetheless, I will try this method and see how well my numbers match up with his.

OK, I’ll post my results piecemeal. I’ll start with outfielders. I’m using the UZR figures from 2002 to 2008, arm/dp ratings included. DG are UZR defensive games. For players with 0 DG (players who played the positions in question but had no UZR chances), I set to 0.25 DG:

LF/RF

DG: 11686
LF_UZR: 80.9
RF_UZR: 13.2
Advantage: RF over LF by 0.87 runs / 150 DG.

CF/Corner OF

DG: 16074
CF_UZR: -404.0
Corner_UZR: 509.0
Advantage: CF over Corner OF by 8.52 runs / 150 DG

Actually some players have 0 DG because they are rounded down, not because they have no chances.

Now an interesting thing happens when you compare dual position SS and 3B. The 3B comes out ahead by a very small margin:

SS/3B

DG: 4600
SS_UZR: -12.5
3B_UZR: -17.0
Advantage: 3B over SS by 0.15 runs / 150 DG

Dual position 3B and 2B come out nearly even:

2B/3B

DG: 6011
2B_UZR: -40.54
3B_UZR: -33.79
Advantage: 2B over 3B by 0.17 runs / 150 DG

So I do feel comfortable in combining 2B and 3B when comparing to SS. But here is SS/2B by itself:

SS/2B

DG: 6255
SS_UZR: -106.9
2B_UZR: 62.7
Advantage: SS over 2B by 4.07 runs / 150 DG

So I guess you can say an approximate 4 run advantage for the SS over 2B/3B, if you cheat and combine 2B and 3B before comparing to SS.

I do have one idea which would account for the aging bias. Compare dual position players only in single seasons and then sum up the single seasons when finished. But you would reduce your sample size somewhat by doing this.

Terps—I have run the numbers on single season matches (not age adjusted) and it is linked to my name.  Results are pretty similar except SS/3B

Jeff,

Thanks for the link. Looking at your numbers, the average difference between a corner OF and CF is 9.6 runs. You give a 2.6 run advantage to RF over LF.

I don’t think you would need to account for aging when comparing single seasons. Well, an age adjustment would probably still be beneficial, but you are going to have less of a bias comparing single seasons than you will if you aggregate the data.

Anyway, I am still thinking of the best way to compare IF/OF.